Enterprise flash storage price drop tops 2019 predictions

Enterprise flash storage predictions

Don Jeanette

Don Jeanette, vice president, Trendfocus: NAND will continue to be in oversupply well through the first half of the year and probably crossing over into the second half. Customers who want to stock up on NAND or SSDs are definitely going to see double-digit percentage takedowns compared to what they were paying last year. It is the Wild West out there on price declines. Enterprise SSDs are pricing out anywhere from 15 cents to 19 or 20 cents a GB at the low end, and that's just going to continue to go down. We were at 30-something, 40-something cents a GB just over a year ago, so that's a significant takedown in one year.

David Floyer

David Floyer, CTO, Wikibon: Flash is going to be the default for pretty much all production systems and will increasingly become the default for a higher percentage of long-term storage. With NAND shipments up and prices coming down, that's going to take away some profits from the vendors. It's going to be a tougher year for them.

Jim Handy

Jim Handy, general director and semiconductor analyst, Objective Analysis: NAND is more affordable than in the past, and the cost of dynamic RAM is about to undergo a very steep collapse, selling at less than one-third of its 2018 peak. This implies that the memory/storage complement of servers, especially in the hyperscale data center, will veer toward SSDs and away from memory in the beginning of the year but then will lean back toward DRAM late in 2019.

QLC flash in 2019

Andy Walls

Andy Walls, IBM fellow and CTO Flash Storage: In 2019, we're going to see the encroachment of flash into areas where HDDs are still very strong. This is going to happen because of QLC flash, with its TCO benefits and its tremendous density. QLC has attributes that might prevent it from entering the enterprise right away, but we're going to see it used in applications where you don't have to access the data much and you don't have to write very often. We're going to see it especially used heavily by the hyperscalers and in areas where 7,200 rpm nearline HDDs were before.

Chris Evans

Chris Evans, director, Brookend: QLC flash will kill off the HDD -- except for archive drives. The economy of scale and increase in capacity of QLC NAND will make QLC more cost-efficient than 15K and 10K spinning media, when compared on a TCO basis. QLC drive capacities will hit 32 TB to 64 TB in 2019, with all of the major vendors getting into the market. This will be double the capacity of even the best HDD, which is currently around 15 TB. With new form-factor devices and server redesigns, QLC will provide for much higher density and more power and cooling efficiency than was possible with spinning media. QLC flash will be a strong contender for edge and IoT use cases where preprocessing of data is important. HDDs that spin at 7.2K or less will be the last refuge for the HDD manufacturers. While the change to QLC won't be completed in 2019, the process will be well and truly underway this time next year.

Marc Staimer

Marc Staimer, president, Dragon Slayer Consulting: Storage tiering is going to make a major comeback this year, between NVMe, SATA and SAS drives; TLC and QLC flash; and SCM. You're going to have storage systems with a variety of SSDs that have different performance and wear characteristics, and it's far easier to tier within a storage system than between storage systems. You're going to have high-performance drives and high-capacity, lower-performance drives, and lower-wear life or read-optimized drives. As data cools, you want to move it off the high-performance, high-wear life drives to lower-performance, higher-capacity, lower-wear life SSDs. It's going to become really important to distinguish between the types of drives in a system.

Jeanette, Trendfocus: There's a lot of talk about QLC NAND hitting the market in 2019, but I don't see that happening in significant volume this year due to the applications that it needs to go into. The current TLC 3D NAND technology has worked and is efficient, whereas the next-generation QLC NAND isn't going to be as reliable or as high performance. You might see cloud vendors take some QLC, but it's going to be more of a situation where they're trying to make it work.

NVMe-oF options

Chadd Kenney

Chadd Kenney, CTO and vice president of product and solutions, Pure Storage: NVMe and, in particular, NVMe-oF, comes into full fruition, getting applications closer to the storage and embedding architectural efficiencies, lower latencies and higher concurrency. We believe RDMA over Converged Ethernet (RoCE) will be the initial deployment option, as DAS latencies can be achieved while enabling the efficiencies of shared storage. After the DAS consolidation of RoCE, FC-NVMe will enable the benefits of NVMe for traditional enterprise applications as support becomes available. NVMe over TCP will bring on everyone else.

Evans, Brookend: NVMe-oF will be big for storage networking -- but only for Fibre Channel. NVMe can be used today with the latest FC devices, so enterprise customers can migrate with little or no impact. This risk-avoidance will make FC-NVMe the initial protocol of choice for end users, rather than ripping and replacing networks with NVMe over Ethernet or InfiniBand. NVMe-oF on Ethernet/Infiniband will see some strong niche use cases where disaggregation or ultrahigh performance is needed.

Scott Sinclair

Scott Sinclair, senior analyst, Enterprise Strategy Group: NVMe-oF adoption will accelerate this year, and NVMe over Fibre Channel will see higher growth than the IP options because of the simplicity of deployment. The Fibre Channel vendors have done an excellent job integrating NVMe-oF capability into their existing product lines, and the technology will roll into IT environments with minimal disruptive impact. While there's innovation on the IP side, there are multiple options, and those considerations will cause some hesitation. Do I want to do RoCE and make an investment in new host bus adapters? Do I want to do NVMe over TCP and wait for support? Each requires different deployment models.

NVMe-based SSDs

Mike Heumann, managing partner, G2M Communications: In 2018, NVMe-based storage devices exceeded the shipments of SAS and SATA storage devices in the enterprise. In 2019, we'll see the NVMe-based U.2 storage form factor achieve 50% of all 2.5-inch flash storage devices shipped in the enterprise market. This is important for enterprise IT professionals because U.2 drives are hot-pluggable and front-accessible and are now supported by the majority of server models from both server OEMs and white-box/ODM server manufacturers.

Steven Hill, senior analyst, 451 Research: Flash will find a growing number of use cases beyond the performance category. For years, the cost of flash over disk has relegated it to high-performance applications. But NVMe, and by extension, NVMe-oF, has the potential to radically shift the price/performance/density model toward all-flash sooner than anyone expected. In 2018 there were several offerings based on the "ruler" NVMe form factor that could eventually pack as much as a petabyte in a single rack unit. That kind of density opens the door for flash to be viable for secondary applications, such as large media repositories and busy data lakes, or as a faster data protection tier for other high-performance storage systems.

Rick Kutcipal, product planning and strategy manager, Broadcom: The availability of high-performance NVMe hardware RAID solutions will accelerate the transition to NVMe drives in enterprise deployments. The main reason there's very little NVMe hardware RAID today is the fact that traditional hardware RAID engines become a bottleneck and don't allow the drive to reach its performance potential. More modern RAID engines that include optimizations targeted at solid-state storage begin to reduce this impact, and significant RAID performance improvements can be achieved.

Emerging memory technologies on the horizon

Walls, IBM: We're going to start seeing SCM deployed in enterprise storage systems, especially on the memory bus. It's still going to be a whole lot more expensive than flash, so it would be a tier in external storage for caching and especially for metadata. In complex storage systems, data reduction and complex data services, you have a lot of metadata. If the metadata is in DRAM or something close to DRAM, you can provide a more consistent response time.

Tim Stammers, senior analyst, 451 Research: We're going to continue to hear about SCM, but the reality is there's only one with the potential for widescale usage. That's the 3D XPoint, jointly developed by Intel and Micron, and used at the moment only in Intel products branded as Optane. Optane products have not sold in large numbers to date, but the emergence of Optane non-volatile memory modules (NVDIMMs) could change that situation. Those NVDIMMs fit only into Intel Cascade Lake processors that aren't generally available yet. But when Intel releases Cascade Lake in volume in 2019, those Optane NVDIMMs could sell a lot better than the Optane NVMe drives that have been in the market.

Greg Wong, principal analyst, Forward Insights: Expect low-latency storage, epitomized by Intel's Optane SSD and Samsung's Z-SSD, to gain further traction in enterprise systems and data centers. The underlying memory technologies -- 3D XPoint and Z-NAND -- have been developed exclusively by Intel and Micron and Samsung, respectively. But I expect competing 3D XPoint- and Z-NAND-like technologies to be announced or introduced in 2019.